alanhuangyoo/crux

187 stars · Last commit 2026-10-07

Taking the pi coding agent to Claude Code-level performance: 0.539 → 0.773 pass@1 on Terminal-Bench 2.1 with the same self-hosted 27B model, by re-engineering the agent.

README preview

<h1 align="center">Crux</h1>

<p align="center">
  <b>A coding agent built on pi, re-engineered for long-horizon terminal tasks.<br/>Terminal-Bench 2.1 pass@1 from 0.539 to 0.773 — on par with Claude Code on the same model.</b>
</p>

<p align="center">
  <img src="https://img.shields.io/badge/Terminal--Bench_2.1-0.773_pass@1-2ea44f" alt="pass@1 0.773">
  <img src="https://img.shields.io/badge/pi_%E2%86%92_Crux-0.539_%E2%86%92_0.773-2ea44f" alt="pi 0.539 to Crux 0.773">
  <img src="https://img.shields.io/badge/Claude_Code%2C_same_model-0.730-555555" alt="Claude Code on the same model 0.730">
  <img src="https://img.shields.io/badge/base-pi-blue" alt="built on pi">
</p>

<p align="center">
  English · <a href="README.zh-CN.md">简体中文</a>
</p>

## Results

<p align="center">

View full repository on GitHub →